A hands-on, resequenced learning platform for the Claude Certified Architect – Foundations certification exam. 30 units across 5 modules, 6 capstone scenario exams, 206 mapped exam concepts.
Built to be worked through with Claude Code, Jupyter notebooks, and interactive HTML overviews.
git clone https://github.com/maxwellsdm1867/caludeArchitect.git
cd caludeArchitect
pip install -r requirements.txt
echo 'ANTHROPIC_API_KEY=sk-ant-...' > .env
./jupyter.sh startOpen notebooks at http://localhost:8888/?token=claude_architect_token and run cells with Shift+Enter.
caludeArchitect/
├── module-1-prompt-engineering/ ← Domain 4 (20%) — 6 units, 47 concepts
├── module-2-context-reliability/ ← Domain 5 (15%) — 6 units, 48 concepts
├── module-3-claude-code-config/ ← Domain 3 (20%) — 6 units, 57 concepts
├── module-4-tool-design-mcp/ ← Domain 2 (18%) — 5 units, 44 concepts
├── module-5-agentic-architecture/ ← Domain 1 (27%) — 7 units, 57 concepts
├── capstone/ ← All domains — 6 scenarios, 42 questions
├── assets/ ← Shared CSS, JS, HTML templates
└── run.py ← CLI dashboard
Modules are intentionally resequenced from the exam guide. The exam lists Agentic Architecture first by weight, but this repo teaches it last because it integrates all other domains.
Each of the 30 units follows 3 stages — fail first, then understand, then prove it:
| Stage | Format | Time | What happens |
|---|---|---|---|
| Card | HTML overview | 2-3 min | Core concept + decision tree + exam trap |
| Lab | Jupyter notebook | 30 min | Wrong approach fails → understand why → build it right |
| Drill | JSON quiz | 5 min | 5 exam-format MC questions per unit |
Plus a challenge with starter code and pytest checker in every unit.
python run.py # Terminal dashboard
python run.py open module-1 unit-1.1 card # Open concept card
python run.py open module-1 unit-1.1 lab # Open lab notebook
python run.py quiz module-1 unit-1.1 # Take the drill| Module | Domain | Weight | Units | Key topics |
|---|---|---|---|---|
| 1. Prompt Engineering | D4 | 20% | 6 | Explicit criteria, few-shot, tool_use schemas, validation-retry, batch API, multi-pass review |
| 2. Context & Reliability | D5 | 15% | 6 | Context degradation, escalation triggers, error propagation, scratchpads, human review, provenance |
| 3. Claude Code Config | D3 | 20% | 6 | CLAUDE.md hierarchy, commands/skills, path rules, plan mode, iterative refinement, CI/CD integration |
| 4. Tool Design & MCP | D2 | 18% | 5 | Tool descriptions, structured errors, tool distribution, MCP servers, built-in tools |
| 5. Agentic Architecture | D1 | 27% | 7 | Agentic loop, coordinator-subagent, context passing, enforcement, hooks, decomposition, sessions |
| Capstone | All | 100% | 6 | Cross-domain scenario exams (customer support, code gen, research, dev productivity, CI/CD, extraction) |
cd caludeArchitect
claudeClaude Code picks up the CLAUDE.md and .claude/ config automatically. With JupyterLab running, Claude Code can edit notebook cells via MCP (NotebookEdit) while you run them interactively in the browser. Use /jupyter in Claude Code to start JupyterLab.
| Detail | Value |
|---|---|
| Format | Multiple choice (1 correct, 3 distractors) |
| Passing score | 720 / 1,000 |
| Scenarios | 4 of 6 picked at random |
| Penalty for guessing | None |
- Programmatic enforcement > prompt-based guidance for compliance-critical rules
- Tool descriptions are the #1 lever for selection reliability
- Few-shot examples beat detailed instructions for consistency
- Context degrades over long sessions — use scratchpads, subagents,
/compact - Subagents don't inherit context — pass it explicitly every time
- 4-5 tools per agent max — more degrades selection
- Independent review instances catch more than self-review
- Batch API has no latency SLA — never use for blocking workflows
MIT